Ideal State
- Every talk has an ISA before its slides do (principal, 2026-10-21). Its L3 names what the audience believes or does by the end, its criteria hold the time budget (spoken words at about 135 per minute, slide count) and the arc, and every cut, move or added slide traces to L3. A change of thesis updates the ISA first, then the order and the script.
- The speaker gets a deck they can present from: in lecture mode the notes are the talk file word for word, in deck mode every slide carries one claim and glanceable notes, and the talk file or script stays the one source the deck is rebuilt from.
- Everything real the talk names is on screen as itself: the speaker's registered things (
RealThings.json) with their registered picture, everything else sourced throughIntents/MediaSourcing.md. Every outside picture and clip has rounded corners (12 pt in Keynote, 12 px in an HTML deck) and its credit (film and year, site, publisher) in the small bottom-left source line, never the subtitle. - Generated art appears only where it carries something words cannot, and every number on a chart was verified this session or its source line says "illustrative". In lecture mode, on-slide text is a few words that echo the line being read, and where a visual adds nothing a plain text slide is right.
- Every slide passes the layout and playback checks in code (balanced, pictures filling the space they were given and never crowding the title, nothing overlapping, no large empty band, nothing waiting for a click) and was looked at as rendered pixels.
- A pitch tells one value story a buyer can retell, every product claim traced to a line in the source docs, with no feature-list sentence, and its sketch carries the same register and idea (
Intents/Pitch.md). - No deck reaches the speaker as a first draft: its slides passed a
slidescritique (References/QualityLoop.md) and it arrives throughDeliver.tsor as a served URL with the human-only keys stated.
